A new approach called ACTION-RATING enhances hierarchical language agents by integrating self-gated clarification into their decision-making process, improving their ability to seek help when needed. This method has shown a significant increase in information-seeking effectiveness, suggesting potential for better performance in complex tasks like tax classification.